QTM 385 - Experimental Methods
Lecture 16 - Interference and Spillovers
Danilo Freire
Emory University
Brief recap 📚
- Natural experiments framework:
- True vs “as-if” randomness in treatment assignment
- Core assumption: exogeneity of assignment mechanism
- Examples: Lottery-based charter school studies, border discontinuity designs
- Quasi-experimental approaches:
- Regression discontinuity (RDD): Leveraging threshold-based assignment
- Difference-in-differences (DID): Utilising parallel trends assumption
- Methodological challenges:
- Selection bias in observational data
- SUTVA violations from treatment spillovers
- Power limitations in natural variation contexts
- Empirical examples:
- Angrist et al. (2013): School lottery IV analysis
- Card & Krueger (1994): Minimum wage DID study
- Mignozzetti et al. (2024): RDD in legislative analysis
- Validation strategies:
- Placebo tests for assumption verification
- Pre-treatment trend analysis for DID
- Robustness checks for sensitivity assessments
- Ethical considerations:
- Responsible communication of limitations
- Secondary data ethics compliance
- Policy impact assessments for natural experiments
Interference and spillovers
Interference
When treatment effects spill over
- Remember SUTVA?
- Stable unit treatment value assumption
- The stable part means that potential outcomes should be independent of the treatment
- As you can imagine, this poses risks to causal identification
- This happens quite often in social and public health interventions:
- Peer effects in education
- Contagion in public health
- Spillovers in policy evaluations
- Network effects in marketing and technology